The Any Vision plugin for Lightroom does a great job hooking up to google,s vision api and getting it to keyword and caption your images.
Giving google some help by passing across image location or gps from the existing metadata, time that the image was taken, etc, significantly improves the results.
The author of any vision has done a great job, within the limitations of the Lightroom plugin interface, but entering the sort of lengthy prompts that give better results becomes painful, if you also want to embed metadata fields in the prompt.
This is exactly the sort of task that ought to be straightforward to do in PM. It certainly ought to be on the roadmap, particularly for the current pricing model. In fact, if I was camerabits, I’d be hiring the anyvision author either as a consultant or to implement the functionality into PM with a better user experience than is possible in Lightroom.